Google's Buyer Direct May Undercut Agentic AI Direct-Sales Startups

Full analysis

Google showed a product at Cannes called Buyer Direct. It lets agency buyers book publisher inventory through one Google interface, skipping the DSP and SSP fee layers. Columnist Gareth Glaser argues this pre-empts the whole crop of agentic AI startups that were pitching software to automate direct-sold ad deals. Since roughly 95% of online publishers already run Google Ad Manager, Google can offer a unified booking rail that outside AI agents, which have to stitch together fragmented publisher APIs, cannot match.

For a technical AI builder, the real question runs past "did Google just win ad-tech." It's whether an agentic AI product with a genuine capability edge can beat an incumbent whose only weapon is that everyone is already logged into their software. This is a Type 1, hard-to-reverse bet if you're a founder mid-build. It's Type 2, easy to revisit, if you're a buyer deciding which vendor to pilot next quarter. The forcing function is the Google DOJ remedy phase, which is live right now and directly targets the exact tying behavior Buyer Direct arguably repeats.

The Skeptic. The 95% GAM stat is carrying more weight than it can hold. GAM being installed is not the same as publishers routing their buy-side relationship through Google after losing an antitrust case over Google's yield behavior. Publisher boards will flag this. For the PM version: Google having the software on every desk does not mean publishers trust Google to also pick their buyers. And "beta at Cannes" is a slide, not a shipped product. The agentic startups aren't dead because a competitor demoed a UI at a conference on the French Riviera. They're at risk because their own pitch, "we glue together publisher APIs," was thin to begin with.

The Safety Lens. The antitrust story and the safety story are the same story. Google was found liable for illegally tying AdX access to GAM. Buyer Direct makes GAM itself a demand source, which is the identical tying logic moved up one layer. The DOJ remedy team is not going to miss that. For the PM version: Google got caught forcing publishers to use one product to access another, and this new thing looks like the same trick with a fresh coat of paint. The systemic risk for anyone building on the open web: one company controls the ad server and the booking interface at 95% penetration, and open-web ad pricing stops being a market and becomes a Google setting.

The Compute Pragmatist. Here is the uncomfortable lesson for AI founders. Buyer Direct almost certainly runs on deterministic business logic with light ML ranking, not on LLM agents burning tokens per deal. The agentic competitors need real-time inventory signals, creative rendering, audience matching, and negotiation inference, and every one of those costs money at inference time. For the PM version: the AI startups pay for a small army of expensive model calls on every transaction, and Google pays for a database lookup. Google's per-transaction cost is an order of magnitude lower. Model quality does not save you when your competitor's marginal cost rounds to zero and yours doesn't.

The Builder. If you're building the "AI replaces the insertion order" play, this is an existential signal, and the correct response is not to rationalize a pivot for next year. GAM is the API you were going to stitch together. Google owns it. Your only live moves are to go where GAM isn't, meaning smaller publishers, audio, out-of-home, CTV through direct integrations, or to build buy-side workflow that Google will not touch precisely because of antitrust optics. For the PM version: your entire moat was doing the plumbing by hand, and the plumbing company just shipped a faucet. The 90-day failure mode is quiet: agency holding companies sit on their hands to watch the Buyer Direct beta, and your demo pipeline dries up while they wait.

The Researcher. The interesting structural question is whether removing the DSP and SSP fee layer actually lifts publisher yield or just moves the margin to Google while feeling frictionless. Auctions need tension. A single booking interface can quietly flatten that tension and suppress price discovery even as it advertises fee savings. For the PM version: cutting out the middlemen looks like it saves money, but if it also kills the bidding competition that set the price in the first place, publishers can end up worse off. The 95% penetration turns Buyer Direct into a natural experiment on whether unified access improves pricing or concentrates it.

Where they part ways

The biggest split is Builder versus Skeptic on whether this is real. The Builder treats Buyer Direct as a shipped moat that ends the agentic-startup thesis. The Skeptic says it's a Cannes demo tested on a beta, and publisher distrust plus a live antitrust remedy could enjoin it before it scales. Both can't be right about the timeline.

Second tension: Compute Pragmatist versus every AI founder's instinct. The Pragmatist says infrastructure access beats model quality in a closed-loop ad system, and the token cost of running real agents is a margin wall Google never hits. That's the bet the agentic startups implicitly made against, and it's the one most likely to sink them regardless of how Buyer Direct's legal fight goes.

Third: Safety Lens versus Skeptic on the regulator. The Safety Lens says the DOJ will read Buyer Direct as the same tying pattern one layer up and move on it. The Skeptic agrees it's vulnerable but bets adoption stalls on its own before litigation catches up. Different mechanisms, same outcome, which is that Buyer Direct may not get to run free.

What this actually hinges on

Three beliefs. One, does 95% GAM installation convert into publishers actually routing buy-side demand through Google after an antitrust loss? Two, is the agentic startups' real problem Buyer Direct, or is it that their per-deal inference cost was always going to lose to deterministic logic? Three, does the DOJ remedy phase constrain GAM before Buyer Direct scales?

The council leans one way clearly. Buyer Direct is a genuine threat to the agentic direct-sales startups, but mostly because those startups had a weak thesis, not because Buyer Direct is a finished product. The access-arbitrage-beats-model-quality point holds whether or not Buyer Direct itself survives. If you're a founder, the thing to verify is whether your demand actually depends on GAM inventory. If it does, you're building on a rail Google owns. If you're a publisher, the clause to watch is whatever the DOJ remedy says about GAM tying, because that determines whether Buyer Direct is even legal at scale.

The call

Prediction: Buyer Direct will still be labeled beta or limited-availability, not generally available across GAM publishers, when Google reports Q4 2026 earnings in early February 2027.

Confidence: Medium. Legal exposure plus publisher distrust slows any GAM-tied launch.

Why: Google was found liable for tying AdX to GAM, and the DOJ remedy phase is live and aimed at exactly that behavior, so shipping a product that makes GAM a demand source is legally radioactive right now. Google's own incentive is to keep this in beta and gather Cannes buzz without handing the remedy team a clean circumvention exhibit. Publishers, already burned once on AdX yield, will move slowly on giving Google the buy-side relationship too, so demand-side adoption won't force Google's hand into a fast GA either. The opposite outcome, a full rollout within six months, would require Google to accept real antitrust risk for a product it can market perfectly well as a beta, which it has no reason to do.

Revisit by 2027-02-10: We're right if Buyer Direct is still gated as beta or limited-availability at Google's Q4 2026 earnings. We're wrong if Google announces general availability across GAM publishers before then.

The founders waiting to pivot should note that the legal drag protecting them is temporary. The cost argument against them is not.